https://github.com/JeeveshN/Facial-Recognition-Tool

:curly_loop: A Facial Recognition tool based on Open-CV.
https://github.com/JeeveshN/Facial-Recognition-Tool

face-detection facial-recognition lbhp opencv prediction

Last synced: 6 days ago
JSON representation

:curly_loop: A Facial Recognition tool based on Open-CV.

• Host: GitHub
• URL: https://github.com/JeeveshN/Facial-Recognition-Tool
• Owner: JeeveshN
• Created: 2016-11-28T18:05:28.000Z (almost 8 years ago)
• Default Branch: master
• Last Pushed: 2020-10-17T09:06:22.000Z (about 4 years ago)
• Last Synced: 2024-08-01T22:49:09.301Z (3 months ago)
• Topics: face-detection, facial-recognition, lbhp, opencv, prediction
• Language: Python
• Homepage:
• Size: 4.09 MB
• Stars: 48
• Watchers: 7
• Forks: 20
• Open Issues: 1
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Face-Recog-Tool

This is a command line tool for facial recognition.It uses image processing library OpenCV.

### Installation

```

git clone https://github.com/JeeveshN/Face-Recog-Tool.git

sudo apt-get install python-opencv

pip install numpy

```

## Getting Started

**Training Set** has to be made before the actual training begins.Multiple pictures of the subjects shot from different angles with preferably different facial expressions would yield the best results.   

Their are 2 ways to achieve this:  

### Fast Mode

Multiple pictures of a single subject are kept in the folder named **Training** (Only single face must be present in each) and 

```

python Train_Faces.py "Name of the Subject"

```

This automatically detects and crops out the face of subject and stores the image in **Dataset** folder (Training Set)

### Learn From Failure

After each prediction the program will ask the user about the wrong predictions and what their correct predictions are supposed to be and will learn and keep getting better accordingly.

Their is a number below each of the detected faces so the user can tell the program which face was wrongly predicted. 

```

Number below face:Correct Name

```

**NOTE**-IF no wrong predictions are made make sure you enter 'Nil' in the prompt for wrong predictions.

### Learning from wrong Prediction

![](/extras/Selection_017.png?raw=True)   ![](/extras/Selection_019.png?raw=True)  

![](/extras/Selection_018.png?raw=True) 

### Underlying Algorithm

The program uses **Linear Binary Pattern Histogram** approach.The basic idea is not to look at the image as a whole but describe local features of object.

To calculate Local Binary Pattern texture descriptor of a pixel we see all of it's neighbouring 8 pixels and mark a 0 if neighbor is less than center or else a 1 and the binary value calculated by moving around is LBP of center pixel.  

For more details refer to [This](http://docs.opencv.org/2.4/modules/contrib/doc/facerec/facerec_tutorial.html#local-binary-patterns-histograms)   

![](/extras/lbp.png?raw=True)  

## Demo

After their are sufficient number of pictures in the training set preferably 25-30 of each subject then we can start to use the program to make predictions.

```

python Face-Recog.py "Image Path or name(if in same folder)"

```

![](/extras/Selection_016.png?raw=True)

![](/extras/2.png?raw=True )

*Note*: Accuracy is low right now because the number of pictures in the training set are less, as the **Dataset** will grow with time predictions would become more and more accurate 

### Naming Convention For Dataset

Each and every picture to be used in the training set must start with the name of the **Subject** followed by any **Number**